Summary
MCP supports prompts as a first-class resource type, separate from tools. Exposing step-by-step workflow guides as MCP prompts gives agents a starting script for common multi-step tasks — without requiring them to discover the correct sequence through trial and error.
Background
Currently agents must infer the correct tool call sequence from tool descriptions alone. For tasks like deploying an HTTP proxy or configuring DNS, the sequence spans 4-6 tool calls across multiple resources. A prompt resource encodes this sequence explicitly and surfaces it via standard MCP prompt listing.
Proposed workflows (initial set)
- Deploy an HTTP proxy — set org → set project → create gateway → create httpproxy → create httproute
- Configure DNS for a domain — set org → set project → create dnszone → create dnsrecordset
- Onboard to a new project — list orgs → set org → list projects → set project → verify context
Acceptance criteria
Ratings
| Dimension |
Score (1–5) |
| LOE |
2 |
| Impact |
4 |
| Agent XP |
5 |
Summary
MCP supports prompts as a first-class resource type, separate from tools. Exposing step-by-step workflow guides as MCP prompts gives agents a starting script for common multi-step tasks — without requiring them to discover the correct sequence through trial and error.
Background
Currently agents must infer the correct tool call sequence from tool descriptions alone. For tasks like deploying an HTTP proxy or configuring DNS, the sequence spans 4-6 tool calls across multiple resources. A prompt resource encodes this sequence explicitly and surfaces it via standard MCP prompt listing.
Proposed workflows (initial set)
Acceptance criteria
Ratings